loyoladatamining/usajobs_validation

USAJOBS Dataset (Validation Sample)

Dataset Description

The USAJOBS Dataset is a comprehensive collection of federal job postings from January 2017 through March 2026. This dataset includes full-text job descriptions, and structured metadata (job title and employer).

This particular dataset presents a sample of sentence-level data from the corpus, tagged with task, skill, and AI attributes.

Dataset Structure

The dataset contains 20k sentences sampled from USAJOBS postings. Each sentence has been marked by an ensemble of three LLMs (GPT-5.4-nano, Gemini-3.1-flash-lite, and Claude-Haiku-4.5) for for three attributes: task, skill, and AI statements. Each of these attributes has a lenient and strict variant, explained below.

Data Fields

Each split contains the following fields:

• —text: (string) Sentence from a USAJOBS posting.
• —task_lenient: (boolean) Whether at least 2/3 LLMs marked the sentence as a task statement.
• —task_strict: (boolean) Whether all 3/3 LLMs marked the sentence as a task statement.
• —skill_lenient: (boolean) Whether at least 2/3 LLMs marked the sentence as a skill statement.
• —skill_strict: (boolean) Whether all 3/3 LLMs marked the sentence as a skill statement.
• —ai_lenient: (boolean) Whether at least 2/3 LLMs marked the sentence as an AI statement.
• —ai_strict: (boolean) Whether all 3/3 LLMs marked the sentence as an AI statement.

Citation

If you find this dataset useful or utilize it for your work, please consider citing the following:

@article{meisenbacher2025extracting,
  title={Extracting O* NET Features from the NLx Corpus to Build Public Use Aggregate Labor Market Data},
  author={Meisenbacher, Stephen and Nestorov, Svetlozar and Norlander, Peter},
  year={2025}
}